A method and system for monitoring crop growth by multi-scale unmanned aerial vehicle image
By combining multi-scale drone imagery monitoring with AI recognition models, the problems of low efficiency and difficulty in accurate diagnosis in large-scale farmland monitoring in existing technologies have been solved, achieving efficient and automatic crop anomaly diagnosis and decision support.
Patent Information
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- 长沙银汉空间科技有限公司
- Filing Date
- 2026-03-05
- Publication Date
- 2026-06-05
AI Technical Summary
Existing technologies cannot simultaneously achieve rapid coverage of large areas of farmland and precise diagnosis of crop anomalies in localized areas, and cannot efficiently and automatically locate and identify stress types such as water shortage, fertilizer deficiency, and pests and diseases.
By employing a multi-scale UAV image monitoring method, and through a two-level operation mode of high-altitude coarse screening and low-altitude fine inspection, combined with UAV image acquisition at different altitudes and AI recognition models, the automated monitoring and anomaly diagnosis of crop growth can be achieved.
It enables rapid coverage of large areas of farmland and accurate diagnosis of suspected anomalies, automatically locates abnormal areas, identifies different stress types and provides precise pesticide and fertilizer application decisions, thereby improving operational efficiency and diagnostic accuracy.
Abstract
Description
Technical Field
[0001] This invention relates to the fields of agricultural remote sensing, drone monitoring and artificial intelligence recognition technology, and more specifically, to a method and system for monitoring crop growth using multi-scale drone images. Background Technology
[0002] Traditional crop growth monitoring relies heavily on manual field inspections, which suffers from inherent drawbacks such as low efficiency, limited coverage, and strong subjectivity in judgment results. Utilizing drones for remote sensing monitoring is a current trend. However, monitoring strategies based on a single flight altitude have inherent contradictions: high flight altitudes can achieve wide-area coverage, but the image resolution is low, making it impossible to identify subtle anomalies at the leaf scale; low flight altitudes, while acquiring high-precision images, have low operational efficiency and are difficult to apply to large areas of farmland. Existing technologies struggle to simultaneously meet the needs of rapid, large-scale screening and precise, localized diagnosis, making it impossible to efficiently and automatically locate abnormal areas and accurately determine specific stress types such as water shortage, fertilizer deficiency, and pests / diseases in large areas of farmland. Summary of the Invention
[0003] The purpose of this invention is to overcome the shortcomings of the prior art and provide a method and system for multi-scale UAV image monitoring of crop growth. Through a two-level operation mode of high-altitude coarse screening and low-altitude fine screening, it can achieve efficient, automatic and accurate monitoring and abnormal diagnosis of crop growth.
[0004] To achieve the above objectives, the present invention adopts the following technical solution:
[0005] A method for monitoring crop growth using multi-scale unmanned aerial vehicle (UAV) imagery includes the following steps: S1. Control the drone to collect low-resolution remote sensing images of the target area at the first altitude; S2. Input the low-resolution remote sensing image into a preset first recognition model to extract the plot boundaries of the target crops; S3. Divide the crop area within the boundary of the plot into multiple grids; S4. Based on the low-resolution remote sensing image, calculate the health index for each grid. S5. Determine the anomaly judgment threshold based on the health index of all grids, mark the grids whose health index meets the anomaly judgment threshold as abnormal grids, and determine the precise location information of each abnormal grid. S6. Control the drone to fly to the precise location of the abnormal grid at the second altitude and collect high-resolution images of the area corresponding to the abnormal grid; S7. Input the high-resolution image into the preset second recognition model and output the abnormal diagnosis result of the crops in the abnormal grid.
[0006] Furthermore, in step S1, the first altitude is 100 meters to 1000 meters; in step S6, the second altitude is 2 meters to 20 meters, and the resolution of the high-resolution image is less than 1 centimeter.
[0007] Furthermore, the calculation method of the health index in step S4 depends on the type of the low-resolution remote sensing image: For RGB images, the health index = (green band value - red band value) / (green band value + red band value); If it is a multispectral or hyperspectral image, the health index = (near-infrared band value - red band value) / (near-infrared band value + red band value).
[0008] Furthermore, the anomaly determination threshold in step S5 is determined based on the statistical values of the health indices of all grids. A preferred approach is to calculate the average and standard deviation of the health indices of all grids, and mark grids with a health index less than (average value - n × standard deviation) as abnormal grids, where n is a preset coefficient, preferably ranging from 1.5 to 3.0.
[0009] Furthermore, the precise location information mentioned in step S5 is the latitude and longitude coordinates of the center point of the abnormal grid, which is used to guide the UAV to accurately re-fly.
[0010] Furthermore, the abnormal diagnostic results described in step S7 include, but are not limited to, at least one of water deficiency, fertilizer deficiency, and pests and diseases. Even further, the second identification model can also be used to identify and output the specific type of fertilizer deficiency (e.g., nitrogen deficiency, phosphorus deficiency, potassium deficiency) or the specific type of pests and diseases (e.g., powdery mildew, rust, aphids).
[0011] Furthermore, as a preferred implementation, steps S2 and S3 are performed during the initial flight, and the extracted plot boundaries and the divided grid information are associated and stored to form a monitoring template; in subsequent flights during the same growing season, the monitoring template is invoked to perform step S3, eliminating the need to repeat plot extraction and grid division, thereby improving operational efficiency.
[0012] The present invention also provides a system for monitoring crop growth using multi-scale unmanned aerial vehicle (UAV) imagery to implement any of the above methods, comprising: The flight control module is used to control the UAV to fly at different altitudes and perform image acquisition tasks at different scales according to instructions; The plot boundary extraction module is used to input low-resolution images into a preset first recognition model to extract the boundaries of crop plots; The grid division module is used to divide the crop area within the boundary of a plot into multiple grids; The health index calculation module is used to calculate the health index for each grid based on low-resolution imagery. An abnormal grid identification and localization module is used to identify abnormal grids based on statistical thresholds and output their precise location information; The refined diagnostic module is used to input high-resolution images into a preset second recognition model and output abnormal diagnostic results for crops.
[0013] Compared with the prior art, the beneficial effects of the present invention are as follows:
[0014] 1. Two-level scale collaborative monitoring: Through the mode of "high-altitude coarse screening + low-altitude fine inspection", it not only ensures rapid coverage of large-scale farmland, but also achieves leaf-level precise diagnosis of suspected anomalies, effectively balancing monitoring efficiency and diagnostic accuracy.
[0015] 2. Fully automated anomaly localization: Based on statistical threshold analysis of the grid health index, the system automatically delineates abnormal areas and generates precise coordinates, eliminating the need for manual interpretation and achieving full automation from data collection to anomaly detection.
[0016] 3. Refined diagnostic capabilities: By combining high-resolution leaf-level images with AI recognition models, it can effectively distinguish different stress types such as water shortage, fertilizer deficiency, and pests and diseases, and can further identify specific species, providing a reliable basis for precise pesticide application, fertilization, and irrigation decisions.
[0017] 4. Operational efficiency optimization: By monitoring the template reuse mechanism (see claim 9), the plot extraction and grid generation steps can be skipped during multiple flights within the same growing season, significantly reducing computational resource consumption and improving operational efficiency. Detailed Implementation
[0018] Example 1: Growth Monitoring Based on Two-Stage Flight of UAV
[0019] This embodiment uses winter wheat growth monitoring as an example to describe the method of the present invention in detail.
[0020] During the winter wheat greening stage, a drone was controlled to fly at an altitude of 300 meters to acquire orthophotos of the target wheat field area, obtaining low-resolution remote sensing images in RGB format with a ground resolution of approximately 10 cm / pixel. The acquired low-resolution RGB images were then input into a pre-trained deep learning segmentation model (such as U-Net). The model automatically identified and segmented all wheat plots in the images, outputting the vector boundaries of the wheat planting areas. Based on the extracted plot boundaries, a 2m × 2m square grid was automatically generated within the boundaries.
[0021] For each grid, iterate through all the pixels it covers, calculate the vegetation index (green-red) / (green+red) for each pixel, and take the average of the vegetation indices of all pixels in that grid as the health index of that grid.
[0022] Calculate the overall mean and standard deviation of the health index for all grid cells. In this example, n=2, and grid cells with a health index less than (overall mean - 2 × standard deviation) are considered abnormal. Extract all abnormal grid cells and calculate the latitude and longitude coordinates of their geometric center points.
[0023] The drone automatically descends to a height of 3 meters and, based on the latitude and longitude coordinates list generated in the previous step, flies sequentially to directly above the center point of each anomaly grid. Using its onboard high-resolution camera, it captures images of the anomaly area at the leaf scale, achieving a resolution of 0.1 cm / pixel. The leaf veins are clearly visible, and the image coverage is slightly larger than the spatial range of the current anomaly grid.
[0024] Each leaf-scale image acquired is input into a pre-trained convolutional neural network classification model (such as ResNet-50). This model has been trained for wheat diseases including water deficiency, nitrogen deficiency, powdery mildew, rust, and healthy leaves. The model outputs diagnostic results, such as "This area is infected with stripe rust, with a confidence level of 92%".
[0025] Example 2: Multiple Flights Based on Monitoring Template Reuse
[0026] In the initial flight of Example 1, the extracted plot boundaries and the divided grid information are associated and stored to form a monitoring template for the wheat field area during this growing season.
[0027] Two weeks later (during the jointing stage), the drone was again used to collect images of the same target wheat field area at the same altitude (300 meters). The drone used RTK positioning technology to ensure that the shooting position and attitude were precisely consistent with the initial flight. After data acquisition, the existing monitoring template was directly invoked, and the new images were automatically divided into 2m x 2m grids corresponding to the template, skipping the steps of plot boundary extraction and grid redrawing. Subsequent steps such as health index calculation, anomaly detection, and low-altitude detailed investigation were the same as in Example 1.
[0028] Example 3: First-level survey based on satellite remote sensing (alternative solution)
[0029] During a monitoring mission, the target area had clear weather with no cloud cover. Recent imagery from the Sentinel-2 satellite (10-meter resolution, including red, green, and near-infrared bands) was used as the first-level low-resolution remote sensing image.
[0030] Satellite imagery is input into the first recognition model to extract crop plot boundaries (consistent with the UAV imagery processing logic). Based on these boundaries, a 10m x 10m grid is created (matching satellite resolution), and the health index NDVI for each grid is calculated as (Near-infrared - Red) / (Near-infrared + Red). The mean and standard deviation of the health index for all grids are calculated. With n=2, grids with a health index less than (mean - 2 × standard deviation) are marked as anomalies, and the latitude and longitude of their center points are calculated.
[0031] Based on the coordinates of the abnormal grid, a drone is dispatched (without the need for a high-altitude survey, directly entering the second stage) to fly to a height of 3 meters above each abnormal point, collect high-resolution RGB images, and input them into the second recognition model to complete the refined diagnosis.
[0032] This embodiment demonstrates the compatibility of the method of the present invention with satellite data sources, as well as the efficient collaborative mode of "satellite survey + drone detailed survey" under clear sky conditions.
[0033] The above description is merely a specific embodiment of the present invention, but the scope of protection of the present invention is not limited thereto. Any variations or substitutions that can be easily conceived by those skilled in the art within the scope of the technology disclosed in the present invention should be included within the scope of protection of the present invention.
Claims
1. A method for monitoring crop growth using multi-scale unmanned aerial vehicle (UAV) imagery, characterized in that, Includes the following steps: S1. Control the drone to collect low-resolution remote sensing images of the target area at the first altitude; S2. Input the low-resolution remote sensing image into a preset first recognition model to extract the plot boundaries of the target crops; S3. Divide the crop area within the boundary of the plot into multiple grids; S4. Based on the low-resolution remote sensing image, calculate the health index for each grid. S5. Determine the anomaly judgment threshold based on the health index of all grids, mark the grids whose health index meets the anomaly judgment threshold as abnormal grids, and determine the precise location information of each abnormal grid. S6. Control the drone to fly to the precise location of the abnormal grid at the second altitude and collect high-resolution images of the area corresponding to the abnormal grid; S7. Input the high-resolution image into the preset second recognition model and output the abnormal diagnosis result of the crops in the abnormal grid.
2. The method according to claim 1, characterized in that, In step S1, the first altitude is 100 meters to 1000 meters; in step S6, the second altitude is 2 meters to 20 meters, and the resolution of the high-resolution image is less than 1 centimeter.
3. The method according to claim 1, characterized in that, The calculation method of the health index in step S4 depends on the type of low-resolution remote sensing image: For RGB images, the health index = (green band value - red band value) / (green band value + red band value); If it is a multispectral or hyperspectral image, the health index = (near-infrared band value - red band value) / (near-infrared band value + red band value).
4. The method according to claim 1, characterized in that, The anomaly determination threshold mentioned in step S5 is determined based on the statistical values of the health index of all grids. Specifically, the average value and standard deviation of the health index of all grids are calculated, and grids with a health index less than (average value - n × standard deviation) are marked as abnormal grids, where n is a preset coefficient.
5. The method according to claim 4, characterized in that, The preset coefficient n is between 1.5 and 3.
0.
6. The method according to claim 1, characterized in that, The precise location information mentioned in step S5 is the latitude and longitude coordinates of the center point of the abnormal grid.
7. The method according to claim 1, characterized in that, The abnormal diagnostic results mentioned in step S7 include at least one of water deficiency, fertilizer deficiency, and pests and diseases.
8. The method according to claim 7, characterized in that, The type of nutrient deficiency or the type of pest or disease is further identified and output by the second identification model.
9. The method according to claim 1, characterized in that, Steps S2 and S3 are performed during the initial flight, and the extracted plot boundaries are associated with and stored with the divided grid information to form a monitoring template; in subsequent flights during the same growing season, the monitoring template is invoked to perform step S3.
10. A system for multi-scale unmanned aerial vehicle (UAV) imagery monitoring of crop growth to implement the method of any one of claims 1 to 9, characterized in that, include: The flight control module is used to control the UAV to fly at different altitudes and perform image acquisition tasks at different scales according to instructions; The plot boundary extraction module is used to input low-resolution images into a preset first recognition model to extract the boundaries of crop plots; The grid division module is used to divide the crop area within the boundary of a plot into multiple grids; The health index calculation module is used to calculate the health index for each grid based on low-resolution imagery. An abnormal grid identification and localization module is used to identify abnormal grids based on statistical thresholds and output their precise location information; The refined diagnostic module is used to input high-resolution images into a preset second recognition model and output abnormal diagnostic results for crops.